Keras监控验证集损失及保存最佳模型

1. EarlyStopping

如果验证集的监控指标不再提升，就让模型停止训练。

from keras.callbacks import ModelCheckpoint, EarlyStopping
custom_early_stopping = EarlyStopping(
    monitor='val_accuracy', 
    patience=8, 
    min_delta=0.001, 
    mode='max'
)

2. 保存最佳模型

model = getModel()
model.summary()

batch_size = 32

earlyStopping = EarlyStopping(monitor='val_loss', patience=10, verbose=0, mode='min')
mcp_save = ModelCheckpoint('.mdl_wts.hdf5', save_best_only=True, monitor='val_loss', mode='min')
reduce_lr_loss = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=7, verbose=1, epsilon=1e-4, mode='min')

model.fit(Xtr_more, Ytr_more, batch_size=batch_size, epochs=50, verbose=0, callbacks=[earlyStopping, mcp_save, reduce_lr_loss], validation_split=0.25)

---

参考：

1. Saving best model in keras;
2. Early Stopping in Practice: an example with Keras and TensorFlow 2.0.